What If the Data Centres Were Drives and the Allies Were the Array?

Pillar 03 · Residency

Bias declaration: I founded SkyeConnex. Weigh my opinion accordingly, then argue with the mechanism.

Every ally is building data centres right now. Canada has a playbook. Europe has about a dozen of them and a committee to sort them out. Australia, the UK and Japan are pouring concrete as fast as they can get power to the sites.

I want every one of them built. Honestly I’d hold the ribbon.

What bugs me is the model underneath. Each country builds a national vault, puts national data in it and locks the door. It’s the Fort Knox theory of the cloud. You can drive by it, put a flag on it, get the minister out there in a hard hat.

Fort Knox works for gold because gold is heavy and nobody needs it on a Tuesday. Data weighs nothing and everybody needs it all the time. Most of its value comes from who else gets to use it with you.

So the vault solves where the data sits. It also creates a pile of new problems.

The vault comes with a bill

Start with latency. Light in fibre does about 200 km per millisecond and it doesn’t care about procurement policy. Keep everything in Canada and our partners in London or Canberra are looking at 80 to 200+ ms on every round trip. Fine for email. Not great for joint ops or anything running inference.

Then concentration. A country’s data sitting in a handful of buildings is a target list. A fire, a flood, a bad firmware push at 4:55 on a Friday, one unhappy contractor. We spent twenty years telling companies not to run on a single site and now we’re building single sites with flags on them.

Then control, which is the one people skip. Sovereign building, sovereign operator, great. Who holds the keys? If the operator can read the data then a court can compel it, an insider can copy it and an attacker can walk out with it. Domestic compelled disclosure is still compelled disclosure. It just shows up in both official languages. Canada’s own security posture now says assume breach. A vault assumes the breach won’t happen.

Add the years it takes to get power, permits, cooling and GPUs, plus every country paying for its own DR site and idle capacity, and you get fifty countries buying fifty copies of the same spare tire.

Allies share. That’s the point

Five Eyes, NATO, the EU, and a few hundred research and defence agreements that nobody has looked at since the signing photo.

Most of that sharing still runs on paper. The treaty says who can see what, then somebody copies the data into the other country’s infrastructure and everyone hopes. Their operators can read it and their courts can compel it. Our control ends when the copy lands.

So countries share less and build more vaults, which makes sharing harder, which justifies the next vault. Self-licking ice cream cone with a security clearance.

What allies actually need is to pool infrastructure without having to pool trust. That’s the problem we built SkyeConnex to solve.

How SkyeConnex works here

Every file gets encrypted inside the owner’s boundary with post-quantum ML-KEM-1024 before it goes anywhere. The keys stay with the owner. No host sees plaintext, and in a sovereign deployment that includes us.

Raidr.cloud, our custody engine, erasure codes the ciphertext into shards. Geopolicy sets the shard count and rebuild threshold per file, so a cabinet document and a cat video don’t get treated the same.

The shards go out to independent hosts in different jurisdictions. We call each host a Connexion. A sovereign data centre in Ottawa, one in Europe, one in Australia, a hyperscaler if you want. Bring your own cloud.

Reassembly only happens inside the client’s boundary.

Now go back to the bill. The client only needs enough shards to hit the threshold, so it pulls from the closest hosts and the array gets faster as it grows. No host holds anything readable, so robbing one gets you a pile of post-quantum confetti. Lose a site and the shards can be rebuilt on another provider from what survived. Serve a court order on any one host and you get shards. Assume breach stops being a slide and becomes how the system behaves.

Every data centre that exists today can be a Connexion today, and the ones finishing next year plug in next year. Nobody waits for the ribbon cutting to get protected. Redundancy gets shared across the alliance instead of rebuilt in every capital.

If a regulator hears “shards abroad” and gets nervous, set geopolicy so a full rebuild quorum stays in Canada. The foreign shards add resilience and reach and no authority at all. Each country sets that for itself.

Treaties turn into settings

Right now a data-sharing agreement is a PDF and a handshake. In SkyeConnex it’s geopolicy. Which jurisdictions hold shards, how many, how many it takes to rebuild, and who holds the keys. Merkle auditing on every shard. The lawyers can still read it, they’re just not the only thing holding it up.

This is the part I keep coming back to. Alliances change. Elections happen, laws get passed at 2 a.m., and a partner you trust in 2026 might be in a very different place by 2029. Nobody says that at the summit dinner.

With a vault, whatever you already copied over is gone and your remedy is a strongly worded letter. With SkyeConnex you disable that Connexion and policy defrag moves the shards to hosts you still trust. Nothing readable ever lived there, so nothing leaks on the way out.

Try this. If you had to cut one ally off from your data tomorrow, how long would it take you today? If the honest answer involves a royal commission, you’re renting your sovereignty.

AI follows the custody

Sovereign AI is mostly being defined as a national GPU farm right now. Same vault, bigger power bill.

SkyeGXU runs training and inference against custody objects. Shards only reassemble inside the client’s boundary. Confidential-computing enclaves that prove themselves with hardware attestation are next. The model loads inside that boundary. Nothing readable crosses a border.

That’s where this goes: allies training on combined data that no single ally would ever be allowed to hold. Health research, threat intel, Arctic and climate data. Everyone keeps custody of their own and the model learns from all of it.

Which country ends up with the better medical model, the one training on its own 41 million people or the one learning safely from close to a billion across its allies? I don’t think it’s close.

Who controls the controller

Any array has a controller, so you should be asking whether SkyeConnex just becomes the new single point of failure. Swap fifty vaults for one company in the middle and you’ve just moved the concentration problem.

Run it sovereign and we never hold keys or plaintext, so compromising us gets you metadata. Escrow is designed to sit with the anchor host in each country under that country’s law. Recovery tooling lets customers pull their shards and rebuild their data without us. And the governance has to fit the mission. A network protecting allied data can’t be one acquisition away from changing passports.

If the controller can be captured it’s just a vault with better marketing.

Start with two countries

You don’t need a treaty to start, which is good because treaties take longer than data centres. One country runs SkyeConnex across its own hosts so no single building holds anything readable. Then two allies connect their sovereign hosts into one array, each with a home quorum. Then a third joins and it’s a network.

So keep building the data centres. Canada needs the capacity. Just build them as drives in an array the whole alliance can stand on.

Ten years from now, are we going to judge a country’s sovereignty by how much data it locked inside its borders, or by how much it could share without losing control of it?

I know where I’d put my money.

Sovereignty by architecture. Location becomes a setting.